[AINode] Decoupling inference manager into request_manager, pool_manager

[yunbow30944]

Description

• Adding pool state and support state transfer
• Decoupling inference manager into request_manager, pool_manager, pool_scheduler
• Adding PoolGroup structure

[Copilot]

Pull Request Overview

This PR implements a major architectural refactoring that decouples the monolithic inference manager into separate, specialized components. The refactoring aims to improve maintainability and separation of concerns by creating distinct managers for different aspects of inference processing.

Key changes:

• Introduces a new RequestManager to handle request lifecycle and result processing
• Adds a PoolManager to manage inference request pools and their states
• Creates a PoolScheduler to handle pool initialization and expansion logic
• Adds pool state tracking with a new PoolState enum and PoolGroup structure

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
inference_manager.py	Refactored to delegate pool and request management to specialized components
request_manager.py	New component managing request lifecycle and result handling
pool_scheduler.py	New component handling pool initialization and expansion logic
pool_manager.py	New component managing pool registration, state tracking, and request dispatch
inference_request_pool_group.py	New data structure grouping pools by model ID
inference_request_pool.py	Added PoolState enum for state management
inference_request.py	Added model_id field to support request routing
decorator.py	Added synchronized decorator for thread safety

[CRZbulabula]

PTAL

[CRZbulabula]

LGTM